Methods for labeling error detection in microarrays based on the effect of data perturbation on the regression model

Area of Science:

• Bioinformatics
• Machine Learning
• Genomics

Background:

• Mislabeled samples in gene expression profiles hinder supervised learning due to disease subtype similarity and misdiagnosis.
• Existing methods like LOOE-sensitivity have limitations in accurately measuring data perturbation effects for mislabeled sample detection.

Purpose of the Study:

• To design a novel method for detecting mislabeled microarray samples by effectively utilizing data perturbation measurements.
• To address the performance limitations of previous data perturbation-based algorithms.

Main Methods:

• Defined a perturbing influence value (PIV) using a support vector machine (SVM) regression model to quantify data perturbation effects.
• Developed Column Algorithm (CAPIV), Row Algorithm (RAPIV), and progressive Row Algorithm (PRAPIV) based on the PIV for mislabeled sample detection.

Main Results:

• All proposed methods (CAPIV, RAPIV, PRAPIV) outperformed the LOOE-sensitivity algorithm on artificial and microarray datasets.
• The progressive Row Algorithm (PRAPIV) demonstrated superior precision and high recall compared to simple SVM and CL-stability.

Conclusions:

• The developed PIV-based algorithms offer a more effective approach to detecting mislabeled samples in microarray data.
• PRAPIV shows significant improvements in detection accuracy, making it a valuable tool for data cleaning in bioinformatics.